Classify text

The Classify text block (AI) uses AI to sort a text input into one or more tags you define.

The Classify text block (AI) uses AI to sort a text input into one or more tags you define. It reads any string, such as a form answer, an email snippet, or a parsed webhook field. It then returns the matching tag rather than the whole record, which is what separates it from Classify record. The block works out the tag but never writes it anywhere. So you must follow it with an Update record or Update list entry block to save the result.

Category	AI
Credit cost	Variable (token-based)
Engine	New Workflows engine (June 2026+)
Plan availability	All plans; AI blocks draw on your monthly workspace credit allowance
Limits	None documented.

Inputs

Input	Type	Required	Default	Notes
Input	Text	Yes	None	The text to classify. Map a trigger field, a parsed JSON value, or an earlier block’s output.
Tags	List of text	Yes	None	The allowed tag options the model may choose from. The model cannot return a tag outside this list.
Allow multiple tags	Boolean	No	Off	When on, the model may return more than one tag for a single input.

Outputs

Output	Type	Example	Notes
Text classification	Text (one or more tags)	Pricing	One of the tags you supplied, or several when Allow multiple tags is on.

Reference the chosen tag in a later block as {{ steps.classify_text.text_classification }}.

Behavior

On each run, the block sends the Input text and your Tags list to the model, then returns whichever tag fits best. The model is limited to your list, so it never invents a category. It always returns something, though, even for vague or off-topic text. A junk input still produces a tag rather than an error.

With Allow multiple tags off, the output is a single tag. With it on, the output can hold several tags. A downstream block must then be ready to receive more than one value, so write the result to a multi-select attribute, not a single-select one.

The block only reads and computes. It does not save its result to any record or list entry. The Text classification output lives in the run until a write block consumes it. Cost is token-based, which means it scales with the amount of text the model reads. Longer inputs and longer tag lists consume more credits per run.

Gotchas

• The result vanishes if nothing saves it. The block computes a tag but writes nothing on its own. Follow it with Update record or Update list entry and map {{ steps.classify_text.text_classification }} into the target attribute.
• A select attribute drops unmatched tags. Writing to a single-select or multi-select attribute requires the option to already exist. Create the matching options on the attribute first, with names that match your Tags values exactly. Otherwise the write silently keeps nothing.
• Overlapping tag definitions make the result unstable. Suppose two tags mean nearly the same thing, like Pricing and Cost. The model then splits ambiguous inputs between them from run to run. Keep tag names mutually exclusive when you need one reliable answer.
• You are paying the model to do a keyword match. Sometimes a fixed rule decides the tag, such as a substring, a sender domain, or a known value. In that case a free Filter or Switch does the same job for zero credits. Reserve this block for genuinely fuzzy text.

Examples

Tagging inbound form answers by pain point

A contact form asks “What’s the main problem you’re hoping to solve?” You want the free-text answer sorted into a fixed set of pain categories before you send the lead to a rep.

Field	Value
Input	{{ trigger.form_answer }}
Tags	Onboarding, Pricing, Integrations, Reporting, Other
Allow multiple tags	Off

For the answer “We can’t get our Salesforce data to sync cleanly,” the block returns Integrations. A following Update record writes that tag to the Primary pain attribute on the person record.

Sorting a parsed webhook field into several themes

A support tool posts ticket text to a Webhook received trigger. You first run Parse JSON to pull out the message field. Then you classify that text into one or more themes. This way a single ticket can carry every relevant tag.

{
  "Input": "{{ steps.parse_json.message }}",
  "Tags": ["Bug", "Billing", "Feature request", "Churn risk"],
  "Allow multiple tags": true
}

For the message “The export button throws an error and honestly we’re reconsidering the renewal,” the block returns Bug and Churn risk. An Update list entry writes both into a multi-select Ticket themes attribute on the support list.

Credit cost worked example

Classify text is priced by tokens. Cost rises with input length and tag-list size rather than a flat per-run figure. Suppose a 60-word inbound form answer against a five-tag list runs roughly 0.3 credits per call, and the form receives 400 submissions a month. That is about 120 credits a month for classification alone. That sits comfortably inside a paid plan’s allowance. Still, confirm it against the exact per-run figure shown in the run viewer before you wire it to a high-volume trigger. Filter out submissions you do not need to classify first. Every record that reaches an AI block costs credits. For plan allowances and how token pricing is billed, see Credits & pricing explained.

Source

Attio Workflows block library. Last reviewed against the Attio Help Center, June 2026.

See also

• Classify record
• Summarize record
• Credits & pricing explained
• AI ICP-fit research and routing
• Attio Workflows block library